> For the complete documentation index, see [llms.txt](https://docs.ta-da.io/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ta-da.io/ta-da-platform/architecture/quality-control.md).

# Quality Control

### Verification flow

Quality controls are made at two levels:

* task level: making sure each task has been properly done
* job level: making sure all tasks have been completed consistently with requirements

Our quality check relies upon 3 levels of validation, both fully customizable based on clients' needs:

* Consensus based verification: verification is done by "community verifiers", and validation is made based on consensus, with "slashing" of incorrect productions or colluding verifiers to incentivize good behaviours
* AI-powered review: specifically trained LLMs help screen compliant tasks to increase verification throughput
* Curator validation: verification is done by specifically trained curators

Depending upon clients' needs and projects' complexity, it is possible to combine any of these 3 verification mechanisms to provide stronger quality checks 

Job validation triggers the distribution of rewards, unless specified otherwise in instructions

### **QC core principles**

Ta-da's platform is build so as to enable quality controls at scale, guaranteeing better data quality, at

* **Redundancy models** (e.g., consensus, majority voting) can be expensive and slow.
* **Automated quality checks** often struggle with edge cases or subtle errors.
* **Data injection** and real-time feedback loops
* **Integration with ML pipelines** or client APIs to close the loop between data needs and data supply.
* **Analytics dashboards** to help data consumers measure dataset evolution, annotation variance, and coverage gaps.